| 1. |
- Gilmore, G., et al.
(author)
-
The Gaia-ESO Public Spectroscopic Survey : Motivation, implementation, GIRAFFE data processing, analysis, and final data products star
- 2022
-
In: Astronomy and Astrophysics. - : EDP Sciences. - 0004-6361 .- 1432-0746. ; 666
-
Journal article (peer-reviewed)abstract
- Context. The Gaia-ESO Public Spectroscopic Survey is an ambitious project designed to obtain astrophysical parameters and elemental abundances for 100 000 stars, including large representative samples of the stellar populations in the Galaxy, and a well-defined sample of 60 (plus 20 archive) open clusters. We provide internally consistent results calibrated on benchmark stars and star clusters, extending across a very wide range of abundances and ages. This provides a legacy data set of intrinsic value, and equally a large wide-ranging dataset that is of value for the homogenisation of other and future stellar surveys and Gaia's astrophysical parameters. Aims. This article provides an overview of the survey methodology, the scientific aims, and the implementation, including a description of the data processing for the GIRAFFE spectra. A companion paper introduces the survey results. Methods. Gaia-ESO aspires to quantify both random and systematic contributions to measurement uncertainties. Thus, all available spectroscopic analysis techniques are utilised, each spectrum being analysed by up to several different analysis pipelines, with considerable effort being made to homogenise and calibrate the resulting parameters. We describe here the sequence of activities up to delivery of processed data products to the ESO Science Archive Facility for open use. Results. The Gaia-ESO Survey obtained 202 000 spectra of 115 000 stars using 340 allocated VLT nights between December 2011 and January 2018 from GIRAFFE and UVES. Conclusions. The full consistently reduced final data set of spectra was released through the ESO Science Archive Facility in late 2020, with the full astrophysical parameters sets following in 2022. A companion article reviews the survey implementation, scientific highlights, the open cluster survey, and data products.
|
|
| 2. |
- Randich, S., et al.
(author)
-
The Gaia-ESO Public Spectroscopic Survey : Implementation, data products, open cluster survey, science, and legacy
- 2022
-
In: Astronomy and Astrophysics. - : EDP SCIENCES S A. - 0004-6361 .- 1432-0746. ; 666
-
Journal article (peer-reviewed)abstract
- Context. In the last 15 years different ground-based spectroscopic surveys have been started (and completed) with the general aim of delivering stellar parameters and elemental abundances for large samples of Galactic stars, complementing Gaia astrometry. Among those surveys, the Gaia-ESO Public Spectroscopic Survey, the only one performed on a 8m class telescope, was designed to target 100 000 stars using FLAMES on the ESO VLT (both Giraffe and UVES spectrographs), covering all the Milky Way populations, with a special focus on open star clusters. Aims. This article provides an overview of the survey implementation (observations, data quality, analysis and its success, data products, and releases), of the open cluster survey, of the science results and potential, and of the survey legacy. A companion article reviews the overall survey motivation, strategy, Giraffe pipeline data reduction, organisation, and workflow. Methods. We made use of the information recorded and archived in the observing blocks; during the observing runs; in a number of relevant documents; in the spectra and master catalogue of spectra; in the parameters delivered by the analysis nodes and the working groups; in the final catalogue; and in the science papers. Based on these sources, we critically analyse and discuss the output and products of the Survey, including science highlights. We also determined the average metallicities of the open clusters observed as science targets and of a sample of clusters whose spectra were retrieved from the ESO archive. Results. The Gaia-ESO Survey has determined homogeneous good-quality radial velocities and stellar parameters for a large fraction of its more than 110 000 unique target stars. Elemental abundances were derived for up to 31 elements for targets observed with UVES. Lithium abundances are delivered for about 1/3 of the sample. The analysis and homogenisation strategies have proven to be successful; several science topics have been addressed by the Gaia-ESO consortium and the community, with many highlight results achieved. Conclusions. The final catalogue will be released through the ESO archive in the first half of 2022, including the complete set of advanced data products. In addition to these results, the Gaia-ESO Survey will leave a very important legacy, for several aspects and for many years to come.
|
|
| 3. |
- Russo, Alessio
(author)
-
Efficient Exploration and Robustness in Controlled Dynamical Systems
- 2023
-
Doctoral thesis (other academic/artistic)abstract
- In this thesis, we explore two distinct topics. The first part of the thesis delves into efficient exploration in multi-task bandit models and model-free exploration in large Markov decision processes (MDPs). This section is driven by the recent research community's interest in uncovering optimal methods to navigate complex decision-making problems. In the second part, we examine attack vectors against MDPs and dynamical systems, which is motivated by the research community's recent push to enhance the safety of Machine Learning models against malicious attacks. Additionally, we explore how to quantify uncertainty in an off-policy evaluation problem, reflecting our ongoing efforts to deepen understanding in this domain. In the first part of the thesis, we present an investigation into the sample complexity of identifying the optimal arm in multi-task bandit problems. In this setting, an agent can select a task and efficiently learn through cross-task knowledge transfer. We derive instance-specific lower bounds that any probably approximately correct (PAC) algorithm should satisfy, revealing both theoretically and empirically significant improvements in scaling over previous methods. Subsequently, we investigate model-free exploration in Reinforcement Learning (RL) problems. By leveraging an information-theoretical viewpoint, we focus on the instance-specific lower bound for the number of samples needed to identify a nearly-optimal policy. We develop an approximation of this lower bound involving quantities that can be inferred using model-free approaches. This leads to a new model-free exploration strategy applicable to continuous MDPs. In the second part of the thesis, we begin by investigating two types of attacks on MDPs: those that alter the observations and those that manipulate the control signals of the victim. We present strategies for designing optimal attacks to minimize the collected reward of the victim. Our strategies show how uncertainties induced by the attack can be modeled using a partially observable MDP (POMDP) framework. We also illustrate how to devise attacks that achieve lower detectability, approaching the problem from a statistical detection perspective. Next, we explore the problem of an eavesdropper aiming to detect changes in an MDP. Approaching this problem from a statistical detection perspective, we introduce a novel definition of online privacy based on the average amount of information per observation of the underlying stochastic system. We derive privacy upper bounds and calculate policies that attain higher privacy levels, supplementing our analysis with examples and numerical simulations. Finally, we present a new off-policy estimation method based on Conformal Prediction that outputs an interval containing the target policy's true reward, demonstrating how to handle the distribution shift between target and behavior policies, and maintain the certainty level while reducing the interval length.Next, we shift our focus onto linear dynamical systems. We study poisoning attacks on data-driven control methods, focusing on how slight changes in the dataset induced by an adversary can significantly deteriorate control methods and potentially destabilize the system. We propose a novel algorithm for computing effective attacks, providing a theoretical basis for our strategy. We also study the detectability of poisoning attacks, focusing on the impact of data poisoning on least-squares estimates. We establish conditions under which the set of models compatible with the data includes the true model of the system, and we analyze different poisoning strategies for the attacker. On the basis of the arguments presented herein, we propose a stealthy data poisoning attack on the least-squares estimator that can evade classical statistical tests.
|
|
